A picture tagged “shield” is findable by “shield.” The reason it isn’t also serving your security, access-control, and hardening pages is that “shield” is one flat label where meaning has several dimensions.
This is the layer above the asset/rendition/tag model of Designing an Image Vault. There, “tag” was a single idea: meaning on the asset. Here we give that meaning structure — because “meaning” is not one list, and treating it as one is what leaves good images stranded.
Table of Contents
One tag is not enough
Tag a security hero simply “shield” and it answers exactly one query: “shield.” It won’t surface for “cybersecurity,” for “access control,” for “the concept of protection,” or for “a technology-domain illustration” — all roles it could fill. A flat label captures one facet of a picture that has several, and every facet you didn’t write is a role the image can’t be found for. Scale that across thousands of assets and most of your library is invisible to most of the searches that should reach it.
Meaning has dimensions: the axes
The fix is to stop treating tags as one bag and split them into axes — independent dimensions of meaning (facets, in classification terms), each answering a different question about the image:
- object — the concrete things depicted (shield, server, prism, tea-cup).
- concept — the abstract idea it serves (cybersecurity, machine learning, thermodynamics).
- context — where or how it’s used (hero, catalog card, social image).
- domain — the field it belongs to (technology, science, business, language).
- visual style — how it looks (vector, gradient, flat-3d; or photographic, cinematic).
- collection — a provenance grouping (which source or track produced it).
- audience — who it’s for.
…and a few more that earn their place only sometimes — action (a depicted activity), status, industry, emotion. A dozen axes is plenty; the point is not the exact list but that each is a separate question, so a search or filter can ask one of them precisely instead of grepping a keyword soup.
A flat tag answers one question; axes let a search ask the right question.
A compute-campus illustration. One flat tag ('server') answers one query. Meaning has several dimensions – so give it structure.
object: server, tower, platform concept: cloud, devops, infrastructure domain: technology, cloud | style: vector context, collection, audience…
Several terms on the concept axis = the same asset surfaces for a cloud page, a devops page, AND an infrastructure page. One asset, 3 roles.
A single UNIQUE key naming THIS asset ('mooc-optics-hero'). Resolves to exactly one image; an exact alias hit outranks every other match.
The other axes find AN image about this; the alias resolves THAT image, precisely.
Give an asset several terms per axis
Within each relevant axis, an asset gets several terms — that’s what widens its reach. A compute-campus illustration tagged object=[server, tower, platform], concept=[cloud computing, devops, infrastructure], domain=[technology, cloud] now surfaces correctly for a cloud page, a devops page, and an infrastructure page — one asset, three roles, because its concept axis names all three. Richer per-axis description isn’t clutter; it’s the difference between an image that serves one page and one that serves the shelf.
The special axis: alias, the exact-match key
One axis behaves unlike the others. The alias is a single, unique term that names this specific asset — the exact-match key a consumer uses to resolve one precise image (a given page saying “my hero is mooc-optics-hero“). Where the other axes are many-to-many and fuzzy — lots of assets share the concept “cybersecurity” — an alias resolves to exactly one asset. Search ranks accordingly: an exact alias hit beats an exact hit on any other axis, which beats a partial match. So the two kinds of lookup coexist cleanly: the alias is addressing (I want that image), the rest of the axes are discovery (find me an image about this). Give every asset exactly one alias, keep it unique, and precise wiring and open search both work off the same tags.
Why multi-axis beats a flat tag bag
Three things fall out of structuring meaning this way that a flat list can’t give you. Ranked search: a match on object or concept is worth more than an incidental term, and axes let the ranker know which is which. Filtering by dimension: “show me everything in the science domain” or “every photographic-style asset” is a clean query against one axis, not a substring guess. And single-sourced meaning: because these tags live on the asset (not per rendition, per the vault model), one structured description travels with every format of the image. The axes turn a keyword pile into something a machine can reason over.
Don’t fill every axis
Structure is a tool, not a chore. You will not populate all dozen axes for every image, and forcing it produces noise that dilutes search. Fill the axes that carry real signal — for content and course heroes that’s typically alias, object, concept, domain, visual style, collection, and context — and leave the rest empty when they’d add nothing. A precise alias plus a few strong object and concept terms is what makes an image both addressable and discoverable; the long tail of axes is there for the images that genuinely need them.
The short version
A single flat tag makes an image findable one way; describing it across axes of meaning makes it reusable across many. Split tags into independent dimensions — object, concept, context, domain, visual style, collection, audience, and a few more — and give each relevant axis several terms, so one asset surfaces for every role its concepts cover. Reserve one axis, the alias, as a single unique exact-match key: the other axes are discovery (find an image), the alias is addressing (resolve that image). Populate the axes that carry signal, not all of them, and the meaning becomes something search can rank and filter rather than grep.
Frequently Asked Questions
Why not just use a flat list of tags?
Because meaning has dimensions. A flat list can’t tell a ranker that an “object” or “concept” match matters more than an incidental word, can’t be filtered by dimension, and tends toward keyword soup. Axes make the same tags answer precise questions.
What are the axes?
Independent dimensions of meaning — object, concept, context, domain, visual style, collection, audience, plus occasional ones like action, status, industry, emotion. The exact set matters less than that each is a separate question you can query on its own.
What makes the alias different from the other axes?
It’s a single unique term that resolves to exactly one asset — the exact-match key for addressing a specific image — whereas the other axes are many-to-many and used for discovery. Search ranks an exact alias hit above everything else.
Do I have to fill in every axis for every image?
No. Fill the axes that carry real signal and leave the rest empty; over-tagging dilutes search. A good alias plus a few strong object/concept terms covers most images.
Where do these tags live — on the file or the image?
On the logical asset, so every rendition (format/size) of the image inherits them. That’s the vault model — see Designing an Image Vault.
